The International Energy Agency (IEA) notes that global data center electricity demand grew by 17 percent in 2025, while consumption from data centers serving AI applications grew 50 percent in the same period.

Meanwhile, governments at international, national, and regional levels continue to tighten sustainability obligations on the industry.

Two of the most pertinent recent examples come from the European Union (EU). In Ireland, data centers with connections of 10 MVA or more must provide matching dispatchable generation or storage and source 80 percent of their annual electricity demand from new renewable generation within the country.

Across the whole of the EU, data centers exceeding 500kW in total capacity must report annually on energy efficiency, water consumption, renewable energy, and waste heat reuse – supporting the EU’s new sustainability rating framework.

As workload profiles diversify, specifying a UPS energy storage system should become increasingly workload-led; AI training, AI inference, high-performance compute (HPC), cloud, storage, and enterprise applications all have different workload profiles and meeting the backup power requirements of each application requires tailored UPS energy storage systems.

Backup power – including UPS energy storage – should be a key component of any data center’s sustainability strategy. However, a workload-led UPS energy storage strategy should shape sustainability strategy – not the other way around.

In other words, rather than fixate on factors like battery chemistry, data centers should focus on how to appropriately address the sustainability challenges of their site’s workload profile.

The diversification of data center workload profiles

AI growth is accelerating data center power demand, but the increasingly common label of ‘AI data center’ conceals significant differences between different AI applications. Training involves processing large datasets to build or refine an AI model, while inference uses a trained model to respond to real-world requests.

These applications create distinct power profiles. Training can involve synchronized power draw across large GPU clusters, with demand fluctuating within milliseconds. Inference is generally more distributed and user-driven, creating bursts of activity across individual servers as requests rise and fall.

The distinction will become increasingly important. JLL forecasts that “AI workloads could represent 50 percent of all data center capacity by 2030,” up from approximately 25 percent in 2025. It also identifies 2027 as a potential inflection point when inference could overtake training as the dominant AI requirement.

The differences between non-AI workload profiles will also remain important, given that non-AI applications could still make up half of all demand by 2030. Cloud platforms, storage environments, and enterprise systems will continue to present different combinations of latency sensitivity, utilization patterns, data availability requirements, and power stability.

How sustainability challenges differ, depending on workload profile

Dynamic workloads

Traditionally, UPS batteries have operated primarily as reserve assets, providing immediate ride-through following a loss of grid power. AI training introduces a different challenge: electrical demand can change at enormous scale and exceptional speed.

Training clusters coordinate thousands of GPUs, causing substantial portions of the load to rise and fall together. Transitions between active computation and checkpointing can occur in milliseconds. The North American Electric Reliability Corporation (NERC) describes one 50MW block of a 200MW training facility in which demand changed sharply over approximately 250 milliseconds, alongside continuing fluctuations during the training run. These characteristics place different demands on electrical infrastructure than the comparatively stable profiles of conventional data centers.

Consequently, some UPS energy storage systems may need to contribute to load smoothing during normal operation, rather than only remaining passive until an outage. This is not simply a matter of increasing battery capacity. It may require an architecture capable of rapid response, high-power discharge, and more frequent cycling. Lithium-ion systems are receiving attention for these applications.

Battery research defines energy efficiency as the ratio between discharged energy and the energy required for charging. Efficiency changes with operating conditions, including temperature, discharge current, and cut-off voltage, and can decline as a battery ages.

Regular cycling also makes energy efficiency a more significant sustainability consideration. Every charge-discharge cycle loses some energy, so the battery must draw more electricity than it subsequently returns.

When cycling is infrequent – such as in systems used primarily for reserve power – these losses may represent a relatively small part of lifetime energy use. However, when storage repeatedly and regularly absorbs and releases power to manage workload fluctuations – such as in the case of dynamic data center workloads – even modest inefficiencies can accumulate, increasing grid consumption and associated Scope 2 emissions.

The sustainability question is therefore not simply whether lithium-ion is used, but how efficiently the complete system will cycle under the workload’s actual power profile throughout its service life.

Reserve power as the primary function

For many data center workloads, UPS energy storage still serves primarily as reserve power.

Even among AI applications, Microsoft research found that although individual servers in AI inference data centers can experience sharp peaks, demand becomes less extreme when aggregated across a cluster, leaving substantially more power headroom than AI training. For many inference environments, UPS energy storage may therefore retain its established role: providing immediate reserve power during an outage rather than routinely supporting workload fluctuations.

A similar principle applies across many non-AI environments, including cloud, storage, and enterprise. Their demand may vary considerably, but it is not generally characterized by the coordinated, cluster-wide GPU power swings associated with large training jobs. Where power demand remains manageable at system level, the priority is reliable standby capacity rather than frequent battery cycling.

However, waiting in reserve does not mean there are no sustainability-related challenges. Lead-acid batteries – still used in a lot of data center reserve power applications – are typically maintained on float charge, meaning a continuous low-level charging voltage is applied to keep them fully charged and ready for use. Because some energy is lost during this process, less-efficient systems require more electricity to maintain readiness. Across a large UPS energy storage installation, these ongoing losses can increase facility electricity consumption and could contribute to Scope 2 emissions.

Sustainability challenges which transcend workload profile

Some sustainability considerations will continue to apply regardless of workload profile.

One is energy density: a more energy-dense system may provide the required backup capacity within a smaller footprint, potentially reducing the quantity of racks, enclosures, cabling, and structural materials required. This can improve overall resource efficiency, particularly where space is constrained. Moreover, greater energy density can reduce the amount of energy required for cooling demand, potentially helping reduce Scope 2 emissions.

Sustainability must also extend beyond operation to the battery’s full lifecycle. The IEA estimates that scaling up critical-mineral recycling could reduce the need for new mining supply by 25 percent to 40 percent by mid-century. It also reports that recycled critical minerals generate, on average, 80 percent fewer greenhouse-gas emissions than primary materials from mining.

Lifecycle responsibility is increasingly becoming a regulatory requirement. The EU Batteries Regulation addresses batteries from material sourcing and carbon footprint through to collection, recycling, and material recovery, including industrial batteries used in energy infrastructure.

As a major battery user, the data center industry must therefore plan for traceability, responsible removal, and recycling from the procurement stage, rather than treating end-of-life management as an afterthought.

Sustainability strategies fit for the data center workload profile

As data center power demands continue to grow rapidly, the challenge for backup power infrastructure is not only the scale of demand, but also the diversification in workload profiles. The industry should take a workload-led approach to UPS energy storage – specifying the UPS energy storage system to fit the workload.

Sustainability challenges can vary, depending on a data center’s workload profile. Dynamic workloads can bring challenges around energy efficiency during charge- and discharge cycles. Meanwhile, UPS energy storage systems used primarily for reserve power can come with energy efficiency challenges around float charging.

The industry cannot follow a UPS energy storage strategy to fit sustainability goals. Instead, data centers should allow their workload-led UPS energy storage system to shape their sustainability strategy.

EnerSys remains committed not only to guiding the industry towards the most effective backup power strategies, but also to helping data centers tackle sustainability challenges, regardless of workload profile.